@thefusioncube

AI Researcher, Neural Nets, and Real Life Agentic Systems.

 San Francisco, CA
Joined November 2024
  • Tweets2.2K
  • Following1.5K
  • Followers1.4K
108Media
Open Sourcing Claude Motion!!! Directly derived from the original system prompts, this allows an agent like Codex and Claude Code to make high quality motion videos! Below is a demo made by GPT 6 Sol within the Chat interface, so it works even with nerfed models.
2
1
7
545
Before it wasn't that bad, but now BE SURE NOT TO COMMIT YOUR .env FILES. xitter.cf/TheHackersNews/status/…
The Hacker News
@TheHackersNews7h
🚨 850+ compromised GitHub accounts. 13,500+ malicious branches. Researchers uncovered 'Gheist,' an ongoing campaign using stolen tokens to plant GitHub Actions workflows that dump private repo secrets into logs. Attackers download the logs, then delete the runs. Details: thehackernews.com/2026/10/gh…
2
1
3
90
I'm so done with cyber. xitter.cf/TheHackersNews/status/…
The Hacker News
@TheHackersNews6h
🚨 UPDATE: The Brotli flaw (CVE-2026-111497) goes beyond Node.js and web traffic. • .NET: patched by Microsoft • FFmpeg: hit via a crafted JPEG XL image • ClickHouse: one request crashes it, fix not yet released • Apache Arrow: no fix yet Details ↓ thehackernews.com/2026/10/br…
1
1
4
424
Ideally, what we want for agents is a holy-grail language. Something that is: 1. As fast and efficient as Rust 2. As little simple as Go 3. Be strictly verifiable. (perhaps with formal proofs) @poteto seems to be taking a first step to make this happen.
20
5
116
13,445
Note: @VictorTaelin was creator, poteto gave a shoutout and used it publicly
1
25
1,338
Sure. That still doesn't change how much the model hallucinates. It's made up fake IPs, pretended to write tests, and just breaks my codebase instead of fixing it. (In Antigravity btw) xitter.cf/wenhaocha1/status/2114…
Wenhao Chai
@wenhaocha111h
Gemini 4 Argon is underrated 😱. When auditing the traces, we found that Pi Harness just ends the run whenever it hits a fallback (e.g. for security task) in 20% of the runs. We rerun Argon with Antigravity Harness, and now it beats Astra and sits right behind Opus 5.5. Congrats to the team @GoogleDeepMind and looking forward to the new ckpt.
3
4
305
X Algo Update TLDR: New posts are no longer cut first when feeds load, giving small creators a fair shot! First 2h & 200 views matter a lot more now: So post originals when followers are online, reply to mutuals & aim for 10s+ reads. Reports/flagging now downweights.
2
6
252
Sweden out here limiting it's own government to the worst AI tool ever to exist. xitter.cf/Polymarket/status/2114…
Polymarket
@PolymarketOct 25
JUST IN: Sweden's government bans ChatGPT & Claude from staff computers, making Microsoft Copilot its only approved AI tool.
27
Dunning-Kruger gives vibe coders wings.
1
1
39
... so it really is evil, huh. xitter.cf/_lyraaaa_/status/21138…
lyra
@_lyraaaa_Oct 24
if you have a qwen3.5 4b around, please try steering neurons L14 N8949 +1 and L13 N7956 -1 now you have evil qwen recipe: dose × median(‖resid‖ leaving layer L, tokens 1:) / ‖W_down[:, n]‖` at the down_proj at every input
2
5
318
Quantization during RL is better than full BF16 RL. Researchers found that quantization for rollouts introduced more exploration, since when quantized, a model is less sure of its decisions. This resulted in both more efficient training and a higher quality model.
1
58
ArXiv paper link: arxiv.org/html/2510.11696v1
25
"The AI bubble is bigger than ever" Opus 5.5, the best and one of the priciest models right now:
1
3
277
I expected Stanford to be first on this list, especially since they take on projects like Marin 535b.
1
67
DROP YOUR SEMI STOCKS EXCEPT FOR CEREBRAS
Jukan
@jukan0515h
Holy smoke NVIDIA has officially lowered the HBM4 pin speed. 11.7-> 9.3Gbps
1
2
427
Lemme hold a Claude max 20x plan!
@jason
@JasonOct 25
Anyone need $4.20 in x money
1
182
But Claude Code definitely has this issue with Opus 5.5! xitter.cf/chaotictransfem/status…
rachel
@chaotictransfem17h
okay claude literally never ends conversations. wtf were you saying to this model?? kinda siding with claude here ngl
1
2
171
That’s literally the whole point of grill-me.
ᴅᴀɴɪᴇʟ ᴍɪᴇssʟᴇʀ 🛡️
@DanielMiessler20h
I disagree with @mattpocockuk that the code should be the source of truth when building with AI. - The intent in our minds is the true source - How we capture that in English is 2nd - Code is an implementation, and it's actually 3rd Writing is closer to thought than code. danielmiessler.com/blog/writ…

Writing Is Closer to Thought Than Code

Why your written intent should be the source of truth when building with AI

danielmiessler.com
1
46
I’m one of them! 16 x RTX PRO 6000 with Prefill-Decode Separation on GLM 5.3! I also do post training and research on the side.
josh avant
@joshavant18h
Right now, there's probably single digit individuals worldwide who are: - highly capable AI devs - invested $1.5M+ in a local inference cluster which can run ~10 fast concurrent agents - using Opus-level open models like Kimi K3.1 ...just living the unrestricted AI maker dream.
2
2
211
AI will find vulnerabilities faster than any Cyber Researcher ever could.
35
Does @elonmusk have Situational Awareness on the topic of neoclouds?
0%Yes
100%No
0%He’s rich, doesn’t matter
1 votes • 10 hours
33
Load more